Advanced Lane Finding¶
Import all required libraries and display calibration image
In [1]:
import numpy as np
import cv2
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import glob
%matplotlib inline
from moviepy.editor import VideoFileClip
from IPython.display import HTML
test_img = cv2.imread('camera_cal/calibration2.jpg')
print("Distorted image")
plt.imshow(test_img)
Out[1]:
Load all calibration images and get the calibration points of of them using cv2.findChessboardCorners
In [2]:
images = glob.glob('camera_cal/calibration*.jpg')
objpoints = []
imgpoints = []
objp = np.zeros((6*9, 3), np.float32)
objp[:,:2] = np.mgrid[0:9, 0:6].T.reshape(-1, 2)
for image in images:
img = mpimg.imread(image)
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
ret, corners = cv2.findChessboardCorners(gray, (9,6), None)
if ret == True:
imgpoints.append(corners)
objpoints.append(objp)
img = cv2.drawChessboardCorners(img, (9,6), corners, ret)
plt.figure()
plt.imshow(img)
Use cv2.calibrateCamera to get the parameters for image undistortion
In [3]:
ret, mtx, dist, rvecs, tvecs = cv2.calibrateCamera(objpoints, imgpoints, gray.shape[::-1], None, None)
dst = cv2.undistort(test_img, mtx, dist, None, mtx)
print("Undistorted image")
plt.imshow(dst)
Out[3]:
Define methods for loading the image and undistoring. Then we run it on all test images to see the results.
In [4]:
test_images = glob.glob('test_images/test*.jpg')
test_images.extend(glob.glob('test_images/straight_lines*.jpg'))
def load_image(img_name):
return cv2.imread(img_name)
def undistort(img):
return cv2.undistort(img, mtx, dist, None, mtx)
undistorted = []
plt.figure(figsize=(12, 28))
i = 1
for timg in test_images:
img = load_image(timg)
dst = undistort(img)
undistorted.append(dst)
plt.subplot(8, 2, i)
i += 1
plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
plt.subplot(8, 2, i)
i += 1
plt.imshow(cv2.cvtColor(dst, cv2.COLOR_BGR2RGB))
Processing pipeline¶
Now that we have calibrated camera and a way to undistort the images we can define a processing pipeline. It contains the following steps:
- Transform the image to birdeye view
- Load the L, B and C channel from the images
- Apply CLAHE, contrast limited adaptive histogram equalization to get more details out of selected channels
- Apply Sobel operator
- Turn the results into binary images
- Combine the binary images and normalize to get the most line coverage possible
- Try to fit lines based on the result:
- Use histogram on the bottom half of the image to find the good starting point
- Move up to find the next part of the line (image is divided into 7 parts) using small windows and checking the found pixel ratios
- Fit a curve based on the found regions
- calculate the line curvature based on the assumed line measurements
- apply drawn lines and measurements onto the processed frame
In [5]:
def l_channel(img):
hls = cv2.cvtColor(img, cv2.COLOR_BGR2HLS)
L = hls[:,:,1]
return L
In [6]:
def b_channel(img):
lab = cv2.cvtColor(img, cv2.COLOR_BGR2LAB)
B = lab[:,:,2]
return B
In [7]:
def v_channel(img):
hsv = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)
V = hsv[:,:,2]
return V
In [8]:
def clahe(img, clip_limit=2.0):
clahe = cv2.createCLAHE(clip_limit, tileGridSize=(8, 8))
return clahe.apply(img)
In [9]:
def binary_image(img, min_treshold, max_treshold):
binary = np.zeros_like(img)
binary[(img > min_treshold) & (img <= max_treshold)] = 1
return binary
In [10]:
def sobel(gray, orient='x', thresh_min=0, thresh_max=255, ksize=3):
if orient == 'x':
abs_sobel = np.absolute(cv2.Sobel(gray, cv2.CV_64F, 1, 0, ksize=ksize))
if orient == 'y':
abs_sobel = np.absolute(cv2.Sobel(gray, cv2.CV_64F, 0, 1))
scaled_sobel = np.uint8(255*abs_sobel/np.max(abs_sobel))
binary_output = np.zeros_like(scaled_sobel)
binary_output[(scaled_sobel >= thresh_min) & (scaled_sobel <= thresh_max)] = 1
return binary_output
In [11]:
def perspective_transform(img):
img_size = (img.shape[1], img.shape[0])
src = np.float32(
[[308, 687],
[1129, 687],
[555, 455],
[735, 455]]
)
dst = np.float32(
[[300, 720],
[1130, 720],
[300, 1],
[1130, 1]]
)
M = cv2.getPerspectiveTransform(src, dst)
Minv = cv2.getPerspectiveTransform(dst, src)
warped = cv2.warpPerspective(img, M, img_size, flags=cv2.INTER_LINEAR)
return warped, Minv
In [12]:
def fit_lanes(binary_warped, nwindows=9, debug=0):
half = binary_warped[binary_warped.shape[0]//2:,:]
histogram = np.sum(half, axis=0)
if debug > 1:
plt.figure()
plt.plot(histogram)
cv2.imwrite('processed/histogram.jpg', histogram)
out_img = np.dstack((binary_warped, binary_warped, binary_warped))*255
midpoint = np.int(histogram.shape[0]/2)
leftx_base = np.argmax(histogram[:midpoint])
rightx_base = np.argmax(histogram[midpoint:]) + midpoint
window_height = np.int(binary_warped.shape[0]/nwindows)
nonzero = binary_warped.nonzero()
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
leftx_current = leftx_base
rightx_current = rightx_base
margin = 80
minpix = 50
left_lane_inds = []
right_lane_inds = []
for window in range(nwindows):
win_y_low = binary_warped.shape[0] - (window+1)*window_height
win_y_high = binary_warped.shape[0] - window*window_height
win_xleft_low = leftx_current - margin
win_xleft_high = leftx_current + margin
win_xright_low = rightx_current - margin
win_xright_high = rightx_current + margin
cv2.rectangle(out_img,(win_xleft_low,win_y_low),(win_xleft_high,win_y_high), (0,255,0), 2)
cv2.rectangle(out_img,(win_xright_low,win_y_low),(win_xright_high,win_y_high), (0,255,0), 2)
good_left_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) &
(nonzerox >= win_xleft_low) & (nonzerox < win_xleft_high)).nonzero()[0]
good_right_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) &
(nonzerox >= win_xright_low) & (nonzerox < win_xright_high)).nonzero()[0]
left_lane_inds.append(good_left_inds)
right_lane_inds.append(good_right_inds)
if len(good_left_inds) > minpix:
leftx_current = np.int(np.mean(nonzerox[good_left_inds]))
if len(good_right_inds) > minpix:
rightx_current = np.int(np.mean(nonzerox[good_right_inds]))
left_lane_inds = np.concatenate(left_lane_inds)
right_lane_inds = np.concatenate(right_lane_inds)
leftx = nonzerox[left_lane_inds]
lefty = nonzeroy[left_lane_inds]
rightx = nonzerox[right_lane_inds]
righty = nonzeroy[right_lane_inds]
left_fit = np.polyfit(lefty, leftx, 2)
right_fit = np.polyfit(righty, rightx, 2)
ploty = np.linspace(0, binary_warped.shape[0]-1, binary_warped.shape[0] )
left_fitx = left_fit[0]*ploty**2 + left_fit[1]*ploty + left_fit[2]
right_fitx = right_fit[0]*ploty**2 + right_fit[1]*ploty + right_fit[2]
y_eval = np.max(ploty)
ym_per_pix = 30/720
xm_per_pix = 3.7/700
left_fit_cr = np.polyfit(lefty*ym_per_pix, leftx*xm_per_pix, 2)
right_fit_cr = np.polyfit(righty*ym_per_pix, rightx*xm_per_pix, 2)
left_curverad = ((1 + (2*left_fit_cr[0]*y_eval*ym_per_pix + left_fit_cr[1])**2)**1.5) / np.absolute(2*left_fit_cr[0])
right_curverad = ((1 + (2*right_fit_cr[0]*y_eval*ym_per_pix + right_fit_cr[1])**2)**1.5) / np.absolute(2*right_fit_cr[0])
if debug > 1:
print(left_curverad, 'm', right_curverad, 'm')
out_img[nonzeroy[left_lane_inds], nonzerox[left_lane_inds]] = [255, 0, 0]
out_img[nonzeroy[right_lane_inds], nonzerox[right_lane_inds]] = [0, 0, 255]
if debug > 1:
plt.figure()
plt.imshow(out_img)
plt.plot(left_fitx, ploty, color='yellow')
plt.plot(right_fitx, ploty, color='yellow')
plt.xlim(0, 1280)
plt.ylim(720, 0)
cv2.imwrite('processed/identified_lines.jpg', out_img)
return ploty, left_fitx, right_fitx, left_curverad, right_curverad
In [13]:
def draw_lines(warped, Minv, ploty, left_fitx, right_fitx, undist, left_curverad, right_curverad):
warp_zero = np.zeros_like(warped).astype(np.uint8)
color_warp = np.dstack((warp_zero, warp_zero, warp_zero))
pts_left = np.array([np.transpose(np.vstack([left_fitx, ploty]))])
pts_right = np.array([np.flipud(np.transpose(np.vstack([right_fitx, ploty])))])
pts = np.hstack((pts_left, pts_right))
cv2.fillPoly(color_warp, np.int_([pts]), (0,255, 0))
newwarp = cv2.warpPerspective(color_warp, Minv, (undist.shape[1], undist.shape[0]))
result = cv2.addWeighted(undist, 1, newwarp, 0.3, 0)
text = "% 6.2f m, % 6.2f m" % (left_curverad, right_curverad)
cv2.putText(result, text, (20, 40), cv2.FONT_HERSHEY_SIMPLEX, 1 ,(255, 255, 255), 2)
return result
In [14]:
def pipeline(img, debug=0):
if debug > 0:
name = img
img = load_image(img)
plt.figure()
plt.title('Original image ' + name)
plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
cv2.imwrite('processed/original.jpg', img)
undist = undistort(img)
if debug > 1:
plt.figure()
plt.title('Undistorted image')
plt.imshow(cv2.cvtColor(img, cv2.COLOR_BGR2RGB))
cv2.imwrite('processed/undistorted.jpg', undist)
transformed, Minv = perspective_transform(undist)
if debug > 1:
plt.figure()
plt.title('Birdeye view')
plt.imshow(cv2.cvtColor(transformed, cv2.COLOR_BGR2RGB))
cv2.imwrite('processed/transformed.jpg', transformed)
l = l_channel(transformed)
if debug > 1:
plt.figure()
plt.title('L Channel')
plt.imshow(l, cmap='gray')
cv2.imwrite('processed/l.jpg', l)
l_clahe = clahe(l)
if debug > 1:
plt.figure()
plt.title('L CLAHE Channel')
plt.imshow(l_clahe, cmap='gray')
cv2.imwrite('processed/l_clahe.jpg', l_clahe)
l_binary = binary_image(l_clahe, 210, 255)
if debug > 1:
plt.figure()
plt.title('L Binary Channel')
plt.imshow(l_binary, cmap='gray')
cv2.imwrite('processed/l_binary.jpg', l_binary*255)
b = b_channel(transformed)
if debug > 1:
plt.figure()
plt.title('B Channel')
plt.imshow(b, cmap='gray')
cv2.imwrite('processed/b.jpg', b)
b_clahe = clahe(b)
if debug > 1:
plt.figure()
plt.title('B CLAHE Channel')
plt.imshow(b_clahe, cmap='gray')
cv2.imwrite('processed/b_clahe.jpg', b_clahe)
b_binary = binary_image(b_clahe, 150, 255)
if debug > 1:
plt.figure()
plt.title('B Binary Channel')
plt.imshow(b_binary, cmap='gray')
cv2.imwrite('processed/b_binary.jpg', b_binary*255)
v = v_channel(transformed)
if debug > 1:
plt.figure()
plt.title('V Channel')
plt.imshow(v, cmap='gray')
cv2.imwrite('processed/v.jpg', v)
v_clahe = clahe(v, clip_limit=6.0)
if debug > 1:
plt.figure()
plt.title('V CLAHE Channel')
plt.imshow(v_clahe, cmap='gray')
cv2.imwrite('processed/v_clahe.jpg', v_clahe)
v_binary = binary_image(v_clahe, 220, 255)
if debug > 1:
plt.figure()
plt.title('V Binary Channel')
plt.imshow(v_binary, cmap='gray')
cv2.imwrite('processed/v_binary.jpg', v_binary*255)
sbl = sobel(l, 'x', 20, 100, ksize=11)
if debug > 1:
plt.figure()
plt.title('Sobel')
plt.imshow(sbl, cmap='gray')
cv2.imwrite('processed/sobel.jpg', sbl*255)
scores = np.zeros(img.shape[0:2]).astype('uint8')
scores += b_binary
scores += l_binary
scores += v_binary
scores += sbl
combined = cv2.normalize(scores, None, 0, 255, cv2.NORM_MINMAX)
if debug > 1:
plt.figure()
plt.title('Combined and normalized')
plt.imshow(combined, cmap='gray')
cv2.imwrite('processed/combined.jpg', combined)
ploty, left_fitx, right_fitx, left_curverad, right_curverad = fit_lanes(combined, debug=debug)
with_lines = draw_lines(combined, Minv, ploty, left_fitx, right_fitx, undist, left_curverad, right_curverad)
if debug > 0:
plt.figure()
plt.title('Original image with lines')
plt.imshow(cv2.cvtColor(with_lines, cv2.COLOR_BGR2RGB))
cv2.imwrite('processed/original_with_lines.jpg', with_lines)
return with_lines
Run the pipeline on a test frame
In [15]:
pipeline('tricky_frames/frame-091.jpg', debug=2);
During the implementation I found a series of tricky frame. Here I run the pipeline to see the results on them.
In [16]:
images = glob.glob('tricky_frames/frame-*.jpg')
for image in images:
pipeline(image, debug=1)
Process the video¶
In [17]:
def process(img):
img = cv2.cvtColor(img, cv2.COLOR_RGB2BGR)
return cv2.cvtColor(pipeline(img), cv2.COLOR_RGB2BGR)
In [18]:
white_output = 'output/project_video.mp4'
clip1 = VideoFileClip("project_video.mp4") #.subclip(22, 24)
white_clip = clip1.fl_image(process)
%time white_clip.write_videofile(white_output, audio=False)
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(white_output))
Out[18]:
In [19]:
white_output = 'output/challenge_video.mp4'
clip1 = VideoFileClip("challenge_video.mp4")
white_clip = clip1.fl_image(process)
%time white_clip.write_videofile(white_output, audio=False)
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(white_output))
Out[19]:
In [20]:
white_output = 'output/harder_challenge_video.mp4'
clip1 = VideoFileClip("harder_challenge_video.mp4")
white_clip = clip1.fl_image(process)
%time white_clip.write_videofile(white_output, audio=False)
HTML("""
<video width="960" height="540" controls>
<source src="{0}">
</video>
""".format(white_output))
Out[20]: